J Behav Med (2015) 38:652–663
DOI 10.1007/s10865-015-9640-7
Exercise habit formation in new gym members: a longitudinal
study
Navin Kaushal1 • Ryan E. Rhodes1
Received: November 12, 2014 / Accepted: March 27, 2015 / Published online: April 8, 2015
 Springer Science+Business Media New York 2015
Abstract Reasoned action approaches have primarily
been applied to understand exercise behaviour for the past
three decades, yet emerging findings in unconscious and
Dual Process research show that behavior may also be
predicted by automatic processes such as habit. The pur-
pose of this study was to: (1) investigate the behavioral
requirements for exercise habit formation, (2) how Dual
Process approach predicts behaviour, and (3) what predicts
habit by testing a model (Lally and Gardner in Health
Psychol Rev 7:S137–S158, 2013). Participants (n = 111)
were new gym members who completed surveys across
12 weeks. It was found that exercising for at least four
bouts per week for 6 weeks was the minimum requirement
to establish an exercise habit. Dual Process analysis using
Linear Mixed Models (LMM) revealed habit and intention
to be parallel predictors of exercise behavior in the tra-
jectory analysis. Finally, the habit antecedent model in
LLM showed that consistency (b = .21), low behavioral
complexity (b = .19), environment (b = .17) and affective
judgments (b = .13) all significantly (p \ .05) predicted
changes in habit formation over time. Trainers should keep
exercises fun and simple for new clients and focus on
consistency which could lead to habit formation in nearly
6 weeks.
Keywords Habit
Dual Process
Exercise
MVPA

Longitudinal
& Navin Kaushal
kaushal@uvic.ca
1 behavior (Sheeran et al., 2013).
Behavioural Medicine Laboratory, Faculty of Education,
University of Victoria, PO Box 3015, STN CSC, Victoria,
BC V8W 3P1, Canada
123
Introduction
Incorporating 150 min of moderate-to-vigorous intensity
physical activity (MVPA) a week has been associated with
the prevention of at least 25 chronic health diseases and
conditions (Garber et al., 2011; Warburton et al., 2007);
however, most people do not meet these recommendations
(Colley et al., 2011; Troiano et al., 2008). Thus, under-
standing factors that contribute to regular MVPA is para-
mount. Research in the past two decades has investigated
this issue primarily through reasoned action approaches
(Hagger, 2010; Head & Noar, 2014; Linke et al., 2014;
Rhodes & Nasuti, 2011), that assume behavior is a volitional
and reflective process (Sheeran et al., 2013). However, a
combination of several recent reviews outlining the short-
comings of reasoned action approaches, combined with
emerging proponents of alternative frameworks, have sug-
gested that a movement beyond reasoned action approaches
could be insightful (Ekkekakis et al., 2013; Rhodes, 2014a,
2014b; Rhodes & Nigg, 2011; Sheeran et al., 2013;
Sniehotta et al., 2014). In line with this reasoning, one di-
rection to consider are models that also incorporate uncon-
scious processes (Sheeran et al., 2013). It has been proposed
that conscious intention and unconscious processes operate
parallel on behavior which is known as a Dual Process ap-
proach (see Evans, 2008 for review). Based on previous
conscious rational models, social cognitive theorists pro-
pose intention to be the strongest predictor of behavior, thus
suggesting intention as the primary conscious motive for
behavior (Ajzen, 1991; Rogers, 1974; Rosenstock, 1974).
By contrast, research in unconscious processes have ranked
habit as possibly the strongest unconscious determinant of
Habit can be defined as ‘‘a learned sequence of acts that
have become automatic responses to specific cues, and are
J Behav Med (2015) 38:652–663
functional in obtaining certain goals or end-states’’ (Ver-
planken & Aarts, 1999, p. 104). Habit is thought to have a
reciprocal relationship with behavior (Gardner, 2014),
where habit affects behavioral repetition but that repetition
also strengthens habit formation. Overall, habit has
demonstrated predictive validity in the physical activity
domain; for example, a recent meta-analysis found it to
correlate r = .43 with behavior which is similar to the
magnitude of the intention-behavior relationship (Gardner
et al., 2011).
Despite the importance of habit outlined in these reviews,
there are still several limitations in the contemporary habit
literature. For example, the majority of the studies on exer-
cise habit are cross-sectional (Gardner et al., 2011). Given
that habit is a dynamic construct, longitudinal studies would
provide stronger support for understanding habit formation
(Gardner, 2014; Lally et al., 2010). To the authors’ current
knowledge only one study has used a longitudinal design to
understand habit development (Lally et al., 2010); the re-
searchers found that it took on average 66 days to develop a
health related habit (healthy eating, drinking and exercise)
among a small student sample. While this is a compelling
finding, it warrants replication and extension with other
samples with a focus on exercise habit. Exercise is a type of
physical activity that is planned, structured, and repetitive
(WHO, 2015). However, it is important to note that 95 % of
adults fail to achieve the recommended physical activity
guidelines (Colley et al., 2011) with the majority of unsuc-
cessful adopters ranking time as the largest barrier to their
exercise (Salmon et al., 2003). With these findings in per-
spective, simply prescribing the general population to exer-
cise every day for over 2 months is not a realistic goal. It
would be helpful to understand the minimum exercise fre-
quency and time required to successfully establish an exer-
cise habit. Behavioral frequency or repetition is a necessary
component for habit formation (Ouellette & Wood, 1998),
thus it would stand to reason that habit formation is partly
dependent on time and frequency. Currently, no study has
examined a time 9 frequency effect on habit formation.
A second shortcoming in the habit and exercise lit-
erature is the limited understanding of the antecedents re-
quired for habit formation. Several models have been
proposed to predict habit formation (Aarts et al., 1997;
Bargh, 1994; Grove & Zillich, 2003; Lally & Gardner,
2013; Triandis, 1977; Verplanken et al., 1997). Despite
some differences in antecedents or the process of habit
formation, these models share the importance of behavior
repetition based on consistent situational cues or context.
One of the most recent models has been proposed by Lally
& Gardner (2013), which suggests that there are four an-
tecedents that are conducive for habit formation: reward,
consistency, environmental cues, and low behavioral
complexity. The researchers theorize the reward compo-
653
nent to be intrinsic, which in exercise research could be
interpreted as positive affective responses to a behavior
(Ekkekakis et al., 2013) or affective judgments (Rhodes
et al., 2009) about the behavioral experience. Affect has
been proposed as having both effects on behavior that are
conscious and unconscious (Custers & Aarts, 2005; Wil-
liams & Evans, 2014; Zajonc, 1980).
Behaviors that are perceived as complex or have not been
sufficiently practiced likely require conscious processes
(Verplanken & Melkevik, 2008; Wood et al., 2002) which
would consequently prevent automaticity. Building from
this research and the habit model proposed by Lally and
Gardner (2013) we theorize that behavioral complexity
represents the level of challenge of performing a task, in-
dependent from motivation or planning. The use of con-
scious process can also be reduced depending on cues present
in the environment. The environment plays a critical role that
can prompt or disrupt automatic behavior (Orbell & Ver-
planken, 2010; Rothman et al., 2009; Wood & Neal, 2009).
Environmental cues, such as mirrors (Sentyrz & Bushman,
1998), lights (Kasof, 2002), or cue cards (Almeida et al.,
2005) have predicted behavior in past research. Additionally,
close proximity to recreation facilities have also been shown
to predict behavior which could act partly via ease of access
but also via environmental cues (Kaushal & Rhodes, 2014;
Moudon et al., 2007; Rhodes, 2006; Rhodes et al., 2007). In
addition to facilitating habit, we theorize that if an individual
does not feel comfortable in a particular environment due to
the presence of any negative cues (i.e. safety concerns, social
physique anxiety, etc.), then the automaticity process would
be interrupted. Hence we theorize that an environment which
provides discomfort functions as a distraction that would
consequently increase the level of conscious awareness and
prevent habit formation.
Consistency is arguably the most unique of the four an-
tecedents as it is a practice rather than feedback (i.e. per-
ceived affect, complexity, or environment). Although the
measurement of temporal consistency in exercise behavior
is scarce, it has been hypothesized that temporal consistency
helps create a protected time for exercise habits (Rhodes &
De Bruijn, 2010). Hence, we define temporal consistency as
performing the behavior at a particular time or after a par-
ticular activity such as exercising regularly at 6 a.m. or after
supper. The closest proposed construct involving consis-
tency is patterned action, i.e. Grove and Zillich (2003).
The purpose of this study was to understand habit for-
mation in new gym members. This was a relevant
population for this study as the enrolment spike during the
New Year followed by a large drop-out of gym members is
a well-known trend but is not clearly understood. The ob-
jectives of the present study were trifold with a focus on
understanding: (1) exercise behaviour, (2) habit formation
and (3) habit predictors.
123
654
1. The first objective was to test the Dual Process ap-
proach by investigating how habit predicts exercise
behavior over 12 weeks while controlling for inten-
tion. It was hypothesized that habit and intention
would both be required to work in synergy to predict
exercise behavior.
2. The secondary objective was to further understand
habit formation by: (1) determining how long it takes
to develop an exercise habit, (2) discerning the cut-off
score for habit, and (3) testing for a time 9 behavior
interaction. The time required for habit formation
would be found by conducting survival analysis. This
analysis would determine when the changes of habit
scores would no longer be significant across time
(Bland & Altman, 1998; Greenhouse et al., 1989; Luke
& Homan, 1998). Habit cut-off score would be re-
vealed by receiver operating characteristic (ROC)
analysis with habit being the test variable and exercise
requirement as state variable; the cut-off score would
be identified from having the highest sensitivity and
lowest specificity values (Greiner et al., 2000; Kraemer
et al., 1999). We hypothesized that habit formation
depends on frequency (Gardner, 2014) and time;
hence, this can be represented with the following
equation: time 9 frequency = habit strength. Previ-
ous analyses which would identify the time required
for habit formation and cut-off score would be sub-
stituted as time and habit strength respectively in the
equation to determine the frequency requirement.
3. The final objective was to test the multivariate model by
Lally and Gardner (2013) to predict habit development.
We hypothesized that habit formation would first de-
pend on affective judgements about exercise, as a re-
peated behavior without reward would require
conscious evaluation. We also expected that complexity
would be a strong antecedent as it could determine if the
behavior is consciously directed or automatically
brought to attention. Finally, we expected that practice
consistency would be a strong predictor of habit for-
mation to reinforce stimulus–response (S–R) (envi-
ronment-affect) as well as operant response (O-R)
(exercise-affect) systems (Skinner, 1954).
Methods
Participants and procedure
One hundred and forty-four adults showed interest in par-
ticipating in our study by requesting a consent form and of
these individuals, 77 % (n = 111) signed the consent and
completed baseline measures. Participants who were exclud-
123
J Behav Med (2015) 38:652–663
ed were those who indicated that they did not meet one of the
following inclusion criteria: (1) being in the age of 18–65, and
(2) being a recent gym member, which was defined as
someone who has joined a gym/recreation centre within the
past 2 weeks. Thirteen gyms and recreation centres were
randomly contacted in the Greater Victoria region in British
Columbia, Canada. Eleven of the thirteen facilities granted
permission to advertise this study. Methods of advertising
included: posting wall posters in high traffic areas (i.e., main
lobby, water fountain, change rooms), placing information
sheets at the main desk, and on-site recruitment which was
performed by the primary investigator. Potential participants
who were interested contacted the primary investigator to
receive the consent form via e-mail along with a web link to
the baseline survey. Consent was implied if participants
clicked on the link and completed the baseline survey. Follow-
up questionnaires were sent at week six, nine, and twelve. We
used a 12 week longitudinal design based on the average time
required to develop habit in a prior research study (66 days)
(Lally et al., 2010). All questionnaires measured the same
constructs described under the Instruments section. The
questionnaires and study protocol were approved by Human
Research Ethics at the University of Victoria.
Instruments
The participants were instructed to consider the definition
of ‘‘exercising regularly’’ as performing 30 min of mod-
erate-to-vigorous in duration five times per week (CSEP,
2012). They were advised to only count exercise that was
done during free time (i.e. not occupation or housework).
Exercise behavior
Exercise behavior was measured by administering the
Godin Leisure Time Exercise Questionnaire (GLTEQ)
(Godin et al., 1986). The questionnaire consists of three
open-ended questions of time and frequency spent on type
of physical activity (mild, moderate and strenuous). The 2-
week test–retest reliability of the measures of total physical
activity and the frequency of activity have been estimated
to be 0.74 and 0.80, respectively (Godin et al., 1986). For
the purpose of this study, only moderate and strenuous
values were used to calculate the exercise behavior. These
categories reflect the definition of moderate-to-vigorous
physical activity provided by recommended guidelines
(CSEP, 2012; Garber et al., 2011).
Exercise habit
Exercise habit was assessed by administering the Self-
Report Behavioral Automaticity Index (SRBAI) (Gardner,
J Behav Med (2015) 38:652–663
2012; Gardner et al., 2012). This scale has been modified
from the Self-Report Habit Index (SRHI) which was de-
veloped by Verplanken and Orbell (2003). The SRBAI
consists of 4 items on a 5-point Likert scale with 1 being
strongly disagree to 5 being strongly agree. The question
stem stated ‘‘When I exercise…’’ which was then followed
by four items on the scale: ‘‘I do it without having to
consciously remember’’, ‘‘I do it automatically’’, ‘‘I do it
without thinking’’, and ‘‘I start before I realize I am doing
it’’. The internal consistencies of this measure were high
across baseline (a = .84), week 6 (a = .92), week 9
(a = .91), and week 12 (a = .95).
Intention
Intention was used as the proximal measure of reflective,
conscious motivation to enact exercise. This construct was
assessed by using a continuous open measure worded, ‘‘I
intend to engage in regular exercise ______ times per week
for the next twelve weeks’’ (Courneya, 1994). Continuous
open measurement of intention preserves scale correspon-
dence with our measure of behavior and has been shown to
be a superior predictor of behavior over dichotomous
closed measures of intention (Courneya, 1994; Courneya &
McAuley, 1994; Rhodes et al., 2006).
Reward
A modified version of the Subjective Exercise Experience
Scale (SEES) (McAuley & Courneya, 1994) was used to
measure exercise reward in the form of affective judgments
about exercise. This instrument has been shown to be a
valid and reliable measure of affect in a variety of exercise
settings (Lox & Rudolph, 1994; McAuley & Courneya,
1994). Items that did not convey a sense of reward were
removed from the scale a priori which were: drain, exhaust,
fatigue, tired, and strong. These terms reflect energy levels
which could be independent from affective reward. For
instance, an individual can experience a very enjoyable run
(intrinsically rewarding) but feel tired after. The remaining
items included: great, positive, terrific, and reverse-scored
items of awful, crummy, discourage and miserable. The
Cronbach alphas across each measurement period were:
baseline (a = .84), week 6 (a = .84), week 9 (a = .86),
and week 12 (a = .90).
Consistency
Temporal consistency had not been assessed in previous
research at the time of the study. Hence, a measure was
created to assess this construct. The item read, ‘‘How
655
consistently did you exercise at the same time each day
(e.g., every morning at 7 am, or exercising daily after
supper)?’’ The options ranged on a 5-point Likert scale
with 1 = not consistent, always at a random time to
5 = very consistent.
Environment
Asking participants to recall an object or context which
functions as a cue has been shown to be problematic
(Gardner & Tang, 2013). The researchers proposed that
individuals may not be able to accurately recall particular
cues as they influence behavioral responses on an uncon-
scious level. With this rationale, it is likely that a distinct
stimuli/change in the environment would be consciously
processed and disrupt automaticity such as encountering a
construction site while driving or the presence of an un-
comfortable object on the driver’s seat. We theorize that an
individual would not be in an automatic state if he/she felt
threatened in the environment as this would trigger con-
scious sensory awareness. Herein, an item worded ‘‘How
comfortable do you feel in your exercise environment’’
which was scored on a five point Likert scale (1 = not very
comfortable to 5 = very comfortable) was used to assess if
the environment supported the process of behavior.
Behavioral complexity
Similar to consistency, a measure to assess behavioral
complexity of performing exercise has not been used in
previous research. A behavior which an individual finds
difficult would require conscious deliberation to perform
and consequently hinder automaticity. The original Self-
Report Habit Index (SRHI) (Verplanken & Orbell, 2003)
recognized this importance and incorporated related items.
The present study applied these items to function as an-
tecedents to automaticity based on the proposed model
(Lally & Gardner, 2013). Hence two items were adapted
from the SRHI which included: ‘‘Exercise is something
that (1) requires effort to do, and (2) I find hard to do’’
(Verplanken & Orbell, 2003). In addition, an individuals’
physical ability could also reflect behavioral complexity.
For instance, a novice exerciser would not be as fluent
exercising compared to an experienced individual. An item
adapted from Rhodes et al., (2006) was also incorporated
into this scale which was worded ‘‘I have good athletic
ability’’. All three items were measured on a 5-point Likert
scale with 1 = strongly disagree to 5 = strongly agree.
The internal consistencies in the present study were:
baseline (a = .80), week 6 (a = .76), week 9 (a = .73),
and week 12 (a = .77).
123
656
Analysis plan
Dual process approach
Linear Mixed Model (LMM) in SPSS 20.0 (IBM, 2011)
was used to understand how intention and habit predicted
exercise behavior across time (Field, 2009; Shek & Ma,
2011; West, 2009). LMM provides strong methodological
advantages over traditional repeated measures analysis of
variance which includes: (1) maintaining precision with
multiple time waves, (2) examining intra- and inter-indi-
vidual differences in the growth parameters (e.g., slopes
and intercepts), (3) selecting an appropriate covariance
structure for the growth curve model (this helps reduce
error variance as researchers can choose the correct model
that reflects the patterns of change over time), and (4)
handling missing data (for further explanation see Field,
2009; Shek & Ma, 2011). In the present study, LMM al-
lowed for simultaneous assessment of the effects of within-
person variation in predictor variables (level 1) across each
time measurement (level 2). Before any analysis was
conducted, the time parameters were grand mean centered
to reduce multicollinearity. The next procedure involved a
series of steps to determine appropriate model fit (Field,
2009). This consisted of first determining if a random in-
tercept would provide a significant difference based on Chi
squared values. A random intercept in a longitudinal model
tests the assumption that each participant can have his or
her own starting point. The next step consisted of calcu-
lating the Intraclass Correlation Coefficient (ICC) on the
baseline model. The ICC describes the amount of variance
in the outcome from differences between individuals. A
high ICC value indicates the stability of the dependent
variable over time. The last step involved conducting a
slope analysis to identify which time polynomial would
provide a suitable fit for the model (Field, 2009; Shek &
Ma, 2011; West, 2009).
Once the model demonstrates appropriate fit parameters
(Field, 2009) then LMM/multilevel analysis can be per-
formed by selecting the Restricted Maximum Likelihood
for estimation method (Field, 2009). Two sets of multilevel
analysis were performed which consisted of testing inten-
tion and habit as predictors of exercise behavior at baseline
and at trajectory/across time.
Habit stabilization, cut-off score, and required frequency
LMM was also used to determine if length of time for habit
formation would be moderated by frequency of behavior.
The interaction can be represented by the following equa-
tion: time 9 frequency = habit strength. Identifying the
values for this equation is a multi-step process which in-
123
J Behav Med (2015) 38:652–663
volved finding how long it takes for habit to develop and
identifying the interaction value. Survival analysis was
used to understand the stability of habit formation; in
particular, this analysis determined when the changes of
habit scores were no longer significant across time (Bland
& Altman, 1998; Greenhouse et al., 1989; Luke & Homan,
1998). The next step involved calculating a cut-off score
for habit formation. Determining the cut-off score was
performed by ROC analysis (Greiner et al., 2000; Kraemer
et al., 1999). ROC curves were constructed by plotting
true-positive rates (sensitivity) against false-positive rates
(1-specificity). ‘‘Habit’’ was the test variable and ‘‘exercise
requirement’’ was the state variable. The cut-off values for
each time period were then averaged to find the overall cut-
off score for the measure. Cut-off values were determined
by identifying points on the curve which demonstrated
maximum sensitivity and minimal specificity. The area
under the curve was also calculated with 95 % confidence
interval (Greiner et al., 2000; Kraemer et al., 1999). Fi-
nally, the time requirement for habit formation and cut-off
values were then substituted as ‘‘time’’ and ‘‘habit
strength’’ respectively in the interaction equation to de-
termine the required minimum ‘‘frequency’’ to achieve
habit formation. This would then be tested by first grouping
participants into meeting, or not meeting the required fre-
quency values then using those groups to predict habit
formation in LMM.
Habit antecedents as predictors of habit formation
LMM was used to test if the antecedents (affect, consis-
tency, complexity, and environment) predicted habit for-
mation. This was a similar procedure to the Dual Process
Approach which first involved testing the antecedents as
predictors of habit at baseline followed by a time-varying
model. Four time measurements of each variable were used
to test if the change of each of antecedent predicted change
of habit in the trajectory LMM.
Results
Descriptives
The mean age of participants was 47.7 (SD = 13.5 years),
70 % were female, and the BMI for the sample was 25.8
(SD = 4.63) suggesting an overweight sample (NIH,
2011). The majority of the participants completed post-
secondary education with 59 % of the sample having a
university degree. Approximately 40 % of the sample had a
household income [$75 000. The participants reported an
average of 186 (SD = 158 min) of total physical activity
J Behav Med (2015) 38:652–663 657

(light, moderate and vigorous) but 72 % were not meeting fit ([.25) and allowed us to proceed with testing indepen-
the recommended exercise guidelines at baseline (CSEP, dent growth curves (Shek & Ma, 2011).
2012). All participants were within their first 2 weeks of
enrolling in their gym or recreation centre and reported Trajectory analysis
being a new member in a gym or recreation facility with
the intention to develop a regular exercise routine. De- Analysis of independent growth curves (IGC) was used to
scriptive data for the participants are displayed in Table 1. understand which polynomial value of time would
Bivariate correlations of the antecedents with habit and demonstrate the best fit for changes in exercise behavior.
exercise are presented in Table 2. The 2-log likelihood was used to calculate the Chi squared
difference which was significant between all three models.
Dual Process approach Since all three time slopes showed significance, the Akaike
Information Criterion (AIC), and Bayesian Information
Model setup and baseline analysis Criterion (BIC) were compared. From these results, the
cubic polynomial was selected as smaller statistical values
Habit and intention were placed in LMM to compare the reflect stronger model fit to the data (Shek & Ma, 2011).
model with and without a random intercept. The Chi The trajectory model found habit habit, b = .23 (p = .001)
squared difference was not significant [v2 (1, N = and intention, b = .23, (p = .007) to be equivalent in
111) = 1.96, p = .37]. Thus, participants’ random starting strength for predicting exercise behavior across time.
points did not significantly change the model (Field, 2009).
The baseline model did not find habit F (1, 101) = .22,
p = .64; or intention F (1, 101) = .90, p = .34 to be a Habit stabilization, cut-off score, and required
significant predictors of exercise. The ICC intercept/(in- frequency
tercept + residual) = .76, suggesting that about 76 % of
total variation from the predictors was due to individual LMM was used to test how frequency and time interacted
differences. ICC values were in acceptable range for model to predict habit. Testing of random intercepts revealed that
the Chi squared difference was not significant [v2 (1,
N = 111) = .06, p = .68]. Thus, a random intercept
Table 1 Descriptive data model did not improve fit (Field, 2009). The baseline habit
Characteristic Percentage
model found habit to significantly predict exercise F (1,
99) = 8.78, p = .004. The ICC value was .38, which
Household income means that 38 % of total variation from exercise was due to
\$50 000 27 individual differences. This was also in the acceptable
$50 001–$75 000 32 range to continue testing IGC.
$75 001–$100 000 19 Test for IGC found a significant Chi squared difference
$100 001–$150 000 12 between linear and quadratic models, v2 (1, N = 111) =
[$150 000 10 14.1, p = .03. It was optimal to proceed with the quadratic
Job status time value for further analysis as: (1) the study consists of
Homemaker 7 four measurement points and a valid polynomial can be a
Temporary unemployed 3 maximum of one less than the number of time points
Part-time employed 23 (Field, 2009), and (2) it has been theorized that habit de-
Full-time employed 51 velops non-linearly (Lally et al., 2010). A quadratic poly-
Retired 16 nomial for time was then used to test habit change across
Education 12 weeks, which was found to be significant F (1,
Less than highschool 1 233) = 14.96, p = .001.
Highschool diploma 23 The next step was to perform Kaplan Meir survivor
College diploma 17 analysis to investigate interaction values at each of the time
University degree 29 slopes. The Kaplan Meir survival curve showed a sig-
Graduate or professional degree 30 nificant difference (p \ .001) between each time curve
Marital status
over the three tests: Log Rank, Brewslow and Trone-Ware.
Never married 13
Each of these tests compares the differences between
Married/common law 76
curves (Breslow = first third of the curve, Trone = middle
section, and Log rank = last third of curve). Pairwise
Separated/divorced/widowed 11
comparisons were used to further determine the significant

123
658
Table 2 Bivariate correlations of habit antecedents with MVPA and Habit
Antecedent Baseline: M, H
Consistency .20*, .48**
Reward .33*, .59**
Behavioural complexity .38*, .38**
Environment cue .19*, .44**
M = MVPA, H = Habit
* p \ .05
differences among the three sections of the curve. This
showed that the second curve (week 6) was significantly
different (p \ .001) than baseline but not with the other
curves (week 9) and (week 12). This stability suggests that
the majority of habit formation in the sample was by week
6 with an interaction value of 12.16 (lower bound of 95 %
Confidence Interval).
Habit cut-off score
Four separate ROC analyses were performed for habit
scores at each time point. The baseline cut-off was 2.91
with a sensitivity of 0.70 and 1-specificity of .18. The
AUC value was 0.76 (95 % CI 0.667–0.858, p \ .001).
Cut-scores for week 6, 9 and 12 were 2.52, 2.76, and 3.01
respectively which averaged a cut-off score of 2.80. The
AUC values ranged from .63 to .76 and were considered in
acceptable range (Akobeng, 2007; Fischer et al., 2003).
Frequency required for habit formation
Previous analysis found habit stabilized at week 6, with the
interaction value of 12.16 (lower bound of 95 % Confi-
dence Interval). We substituted this value with the habit
cut-off score of 2.8 in the equation to solve for minimal
frequency of exercise bouts required to achieve habit for-
mation and found that a frequency of approximately 4 days
per week was required to achieve an interaction score of
12. This finding was tested by first determining if behav-
ioral frequency predicted habit across time. The LMM
analyses found behavioral frequency to predict habit over
12 weeks (b = .24, p \ .001). The next step involved
separating values based on high frequency (C4 days/week)
and low frequency (\4 days/week) groups. When these
groups were then tested as predictors of habit, the low
frequency group did not predict habit (b = .09, p = .42)
but the high frequency group was significant (b = .24,
p \ .001). A descriptive plot was produced to depict how
the frequency groups affected habit scores across time
(Fig. 1). The figure shows that those in the high frequency
group demonstrated stability of habit scores and mainte-
nance of habit (C3/5) scores across the 12 weeks. In par-
123
J Behav Med (2015) 38:652–663
Week 6: M, H Week 9: M, H Week 12: M, H
.20*, .30** .10, .48* .27*, .31**
.29**, .46** .14, .21* .12, .26**
.25*, .63** .27**, .48** .22*, .64**
.26**, .23* .17, .18 .28**, .23*
ticular, week 6 shows that 61.5 % of participants achieved
habit in the high frequency group compared with 44.8 % in
the low frequency. By week 12, the values for high and low
frequency groups were 63.8 and 22.6 % respectively.
Habit antecedents as predictors of habit formation
Baseline
The four baseline antecedents (affect, consistency, complexity,
cues) were placed in LMM to compare two variations of the
model: with and without a random intercept. The Chi squared
difference was not significant [v2 (1, N = 111) = 3.82,
p = .12]. Thus, a random intercept model did not improve fit
(Field, 2009). The ICC intercept/(intercept + residual) = .64,
suggesting that about 64 % of total variation from the an-
tecedents was due to individual differences. LMM analysis of
the baseline habit model found that affective judgments (re-
ward) predicted habit with a medium-large effect size b = .47,
F (1, 106) = 31.56, p \ .001 followed by consistency
b = .45, F(1, 106) = 13.36, p = .001; behavioral complexity
and cues were not significant (Table 3).
Trajectory analysis
The following trajectory analysis would reveal if the an-
tecedents contributed a significant change to habit scores across
12 weeks. A quadratic polynomial was used for the trajectory
analysis as the results from the previous IGC found this time
slope to be a suitable fit for a model with habit as the DV. When
time was added in the trajectory analysis, consistency demon-
strated the largest effect size for predicting habit formation,
(b = .21, p \ .001), followed by low behavioral complexity
(b = .19, p \ .001), environment (b = .17, p = .008) and
affective judgements (b = .13, p = .003) (Table 3).
Discussion
The primary purpose of this study was to understand the
process of habit formation in new gym members over
12 weeks. The secondary purpose was to investigate how
J Behav Med (2015) 38:652–663 659

3.2 significant predictors over time with equal effect sizes

(b = .23, respectively) in support of our hypothesis. The
3
non-significant finding of intention at baseline could be
2.8 attributed to the sample being new gym members with
Habit Score

already high intentions. Intention-based approaches have

2.6
2.4
2.2
2 exercise over the 12 weeks. Overall, the results add support
Baseline Week 6 Week 9
Time
High Frequency
Fig. 1 Habit scores between high and low frequency groups. Note
high frequency (C4 times/week), low frequency (\4 times/week)
the Dual Process approach predicts exercise behavior and
how the antecedents in the habit model predict habit for-
mation. The present study found the SRBAI (Gardner et al.,
2012) to have a cut-off score of 2.80/5. With regards to
behavioural requirement for habit formation, it was found
that participants who exercised for at least four bouts per
week for 6 weeks successfully established an exercise
habit. Dual Process tests showed that intention and habit
were not significant at baseline but they became equal
predictors of exercise behavior in the trajectory analysis.
Finally, the habit model found that affect and consistency
were the largest predictors for people starting a habit;
however, trajectory analyses revealed that consistency was
the most important predictor followed by low behavioral
complexity, environment, and affect.
It was hypothesized that habit and intention would both
be significant predictors of exercise behavior, commensu-
rate with the Dual Process approach. The present study did
not find habit or intention to significantly predict exercise
behavior during baseline; however, both constructs became
been criticised in this particular situation and it represents a
practical application of the intention-behavior gap (Rhodes
& De Bruijn, 2013). However, as time progressed, the
change of intention and habit scores predicted change of
Week 12 to a small literature on Dual Process approach applied to
exercise behavior (Calitri et al., 2009; Conroy et al., 2010;
Low Frequency
Hyde et al., 2010).
In terms of the time required to establish an exercise
habit, the present research found that exercise habit pla-
teaued on the 6th week (42–49 days) of the study with
48 % of the sample achieving habit formation. Previous
work has found that it took an average of 66 days to
establish a health related habit (Lally et al., 2010). How-
ever the differences in methodologies do not warrant much
comparison. For instance, Lally et al., (2010) used a
combination of data and projected analysis to determine
exercise habit formation. The present study also found the
cut-off score of the SRBAI to be 2.8/5 using ROC analyses.
This indicates that 2.8/5 is the minimal score to detect that
the behavior is not entirely controlled by conscious pro-
cesses. Scores C2.8/5 would suggest that automaticity is
significantly involved in the behavior. The score is fairly
low on the measure, suggesting that automaticity may be a
continuum where low scores still represent predictive val-
ues. Scores that are very low on this continuum would
reflect high cognitive process with minimal automaticity
(i.e. controlling air traffic) and the other end of the con-
tinuum would indicate the opposite (i.e. sleeping). These
findings and theorizing satisfy both perspectives of habit
research; the results support theorizing that exercise is not

Table 3 Baseline and trajectory analysis: antecedents as predictors of habit formation

Source B b SE 95 % Confidence lower bound Interval upper bound

Baseline
Consistency .22** .45** .13 .14 .64
Reward .36*** .47*** .08 .30 .64
Complexity .03 .02 .08 -.18 .13
Environment cue .16* .09 .08 -.06 .25
Trajectory analysis
Consistency .26*** .21*** .07 .11 .38
Reward .05** .13** .04 .07 .25
Complexity .09 * .19** .05 .01 .32
Environment cue .14** .17** .07 .08 .36
b = standardized beta, nv = non-significant variability in sample to predict trajectory change
*** p \ .001, ** p \ 0.01, * p \ .005

123
660
completely automatic (Maddux, 1997) yet it demonstrates
that habit may be critical for exercise continuance (Rhodes
& De Bruijn, 2013).
Although, the present study estimated a similar time
required for habit formation to Lally et al. (2010), we also
hypothesized that time would be moderated by perfor-
mance frequency. The results clearly supported this con-
jecture, with a time 9 frequency interaction. A large drop
(44.8–22.6 %) in habit was noticed from week 6 and to 12
in the low frequency group (\4X/week); however, those in
the high frequency group maintained habit across time
(61.5–63.8 %). Theoretically, this pattern aligns with sev-
eral models that propose establishing a habit requires re-
peated behavioral practice across time (Hall & Fong, 2007;
Ouellette & Wood, 1998; Rhodes & De Bruijn, 2013;
Triandis, 1977; West, 2006). Fortunately, these findings are
also aligned with public health guidelines suggesting that
an exercise habit can be achieved in 4–5 bouts with 30/
40 min per session.
We also hypothesized that habit formation would de-
pend on the presence of the antecedents theorized by Lally
and Gardner (2013), with affective judgments and com-
plexity predicting habit in the initial phases but consistency
predicting habit formation over time. We had some support
for this hypothesis. Affective judgments about the exercise
experience were found to be the primary predictor of habit
formation at baseline but consistency became the strongest
predictor in the trajectory analysis. This supports prior
theorizing on the foundation of habits. Affect has been
investigated in understanding general unconscious goals
(Custers & Aarts, 2005) and habit of fruit consumption
(Wiedemann et al., 2014) but not for exercise behavior. It
is likely that negative feelings which stem from un-
favourable experiences could prompt conscious delib-
eration for the individual before performing the behavior.
On the other hand, a positive reward would not require
evaluative process; the presence of positive affect may
drive behavior at an unconscious level (Custers & Aarts,
2005; Zajonc, 1980).
In terms of consistency, our results support our con-
jecture that it may be a pillar in establishing both the sti-
mulus–response (S–R) (environment-affect) as well as
operant response (O-R) (exercise-affect) conditions as the
behavior becomes more familiar. The significant effect of
consistency also helps establish a potentially different an-
tecedent for habit formation than motivation. This con-
struct suggests that how, rather than why one practices may
be more important to forming habits. Hence, these results
suggest that initiating an exercise routine that is enjoyable
and consistent can help in habit formation.
Behavioral complexity was found to predict change of
habit across time which aligns with previously theorized
123
J Behav Med (2015) 38:652–663
research on the importance of low cognitive load for habit
formation (Verplanken & Melkevik, 2008; Wood et al.,
2002). Although exercise is a complex behavior, it was
likely that practicing consistently eventually eased the
challenges of the behavior across time thus allowing for the
facilitation of habit. A comfortable environment that does
not stimulate more conscious thinking was also shown to
predict habit over time. Environment cue assessment from
traditional methods may not be clear due to variability in
the type of cues and the method of measurement (Gardner
& Tang, 2013). Hence the present finding could provide a
novel approach to assess if the environment supports the
development of habit.
Despite the longitudinal design, analyses, methods, and
novel approach to understanding habit and its antecedents,
the present study still has limitations that are important to
address. For instance, although the sample consisted of new
gym members, there was some variability in their exercise
history. Since the present results found habit formation to
occur by week six, this suggests that the majority of var-
iation of habit occurred within this period. Assessing habit
scores more frequently within the first 6 weeks could
provide a more detailed scope of the habit formation phase.
Second, the habit model proposed by Lally and Gardner
(2013) presents a strong case of four antecedents of habit
which have individually been found to correlate with habit
in various studies (Gardner et al., 2011). However, the
authors did not provide suggestions on measuring these
predictors. The present study used a mixture of previous
validated scales and customized items to this model.
Although these scales demonstrated to predict change of
habit across time, other measurements of these constructs
may yield different findings and this warrants sustained
research. Finally, future research should also provide ob-
jective measurement to yield a stronger interpretation of
exercise behavior and habit formation.
In summary, the study found support for the Dual Pro-
cess approach as intention and habit both predicted exer-
cise behaviour over time. Exercising for at least four times
per week for approximately 6 weeks was required to
establish an exercise habit. Although affect was found to be
the strongest predictor at baseline, consistency was the
most important factor for predicting changes in habit. The
environment and low behavioral complexity demonstrated
to play a significant role in changing habit across time.
Exercise promoters should focus on setting a consistent
exercise schedule and keeping the workouts fun and skill
appropriate to increase the likelihood of habit formation. In
addition, the environment should be comfortable and
welcoming for new clients. The first 6 weeks appear cri-
tical for habit formation and new exercisers should strive to
workout at least four times per week.
J Behav Med (2015) 38:652–663
Acknowledgments We would like to thank two anonymous re-
viewers for providing valuable and insightful suggestions to improve
this paper. We would also like to thank Alison Quinlan for her
feedback on the manuscript.
Conflict of Interest Navin Kaushal and Ryan E. Rhodes declare
that they have no conflicts of interest in the research.
Human and Animal Rights and Informed Consent The present
study was approved by the Human Research Ethics Board at the
University of Victoria. All procedures followed were in accordance
with the ethical standards of the responsible committee on human
experimentation (institutional and national) and with the Helsinki
Declaration of 1975, as revised in 2000 (5). Informed consent was
obtained from all participants for being included in the study.
References
Aarts, H., Paulussen, T., & Schaalma, H. (1997). Physical exercise
habit: On the conceptualization and formation of habitual health
behaviours. Health Education Research, 12, 363–374.
Ajzen, I. (1991). The theory of planned behavior. Organizational
Behavior and Human Decision Processes, 50, 179–211. doi:10.
1016/0749-5978(91)90020-t
Akobeng, A. K. (2007). Understanding diagnostic tests 3: Receiver
operating characteristic curves. Acta Paediatrica, International
Journal of Paediatrics, 96, 644–647.
Almeida, F. A., Smith-Ray, R. L., van den Berg, R., Schriener, P.,
Gonzales, M., Onda, P., & Estabrooks, P. A. (2005). Utilizing a
simple stimulus control strategy to increase physician referrals
for physical activity promotion. Journal of Sport & Exercise
Psychology, 27, 505.
Bargh, J. A. (1994). The four horsemen of automaticity: Awareness,
Intention, Efficiency, and Control in Social Cognition. In R.
S. Wyei Jr & T. K. Srull (Eds.), Handbook of social cognition
(2nd ed., Vol. 1, pp. 1–40). Hillsdale, NJ: Erlbaum.
Bland, J. M., & Altman, D. G. (1998). Statistics notes: Survival
probabilities (the Kaplan–Meier method). British Medical Jour-
nal, 317, 1572.
Calitri, R., Lowe, R., Eves, F. F., & Bennett, P. (2009). Associations
between visual attention, implicit and explicit attitude and
behaviour for physical activity. Psychology & health, 24,
1105–1123.
Canadian Society of Exercise Physiology (CSEP). (2012). Canadian
physical activity guidelines and Canadian sedentary behaviour
guidelines. http://www.csep.ca/english/view.asp?x=804.
Colley, R. C., Garriguet, D., Janssen, I., Craig, C. L., Clarke, J., &
Tremblay, M. S. (2011). Physical activity of canadian adults:
Accelerometer results from the 2007 to 2009 canadian health
measures survey. Health Reports, 22, 7–14.
Conroy, D. E., Hyde, A. L., Doerksen, S. E., & Ribeiro, N. F. (2010).
Implicit attitudes and explicit motivation prospectively predict
physical activity. Annals of Behavioral Medicine, 39, 112–118.
Courneya, K. S. (1994). Predicting repeated behavior from intention:
The issue of scale correspondence. Journal of Applied Social
Psychology, 24, 580–594. doi:10.1111/j.1559-1816.1994.
tb00601.x
Courneya, K. S., & McAuley, E. (1994). Factors affecting the
intention-physical activity relationship: Intention versus expec-
tation and scale correspondence. Research Quarterly for Exer-
cise and Sport, 65, 280–285.
661
Custers, R., & Aarts, H. (2005). Positive affect as implicit motivator:
On the nonconscious operation of behavioral goals. Journal of
Personality and Social Psychology, 89, 129–142.
Ekkekakis, P., Hargreaves, E. A., & Parfitt, G. (2013). Invited guest
editorial: Envisioning the next fifty years of research on the
exercise-affect relationship. Psychology of Sport and Exercise,
14, 751–758.
Evans, J. S. B. T. (2008). Dual-processing accounts of reasoning,
judgment, and social cognition. Annual Review of Psychology,
59, 255–278.
Field, A. P. (2009). Discovering statistics using SPSS (3rd ed.).
London: Sage publications.
Fischer, J. E., Bachmann, L. M., & Jaeschke, R. (2003). A readers’
guide to the interpretation of diagnostic test properties: Clinical
example of sepsis. Intensive Care Medicine, 29, 1043–1051.
Garber, C. E., Blissmer, B., Deschenes, M. R., Franklin, B. A.,
Lamonte, M. J., Lee, I. M., & Swain, D. P. (2011). Quantity and
quality of exercise for developing and maintaining cardiorespi-
ratory, musculoskeletal, and neuromotor fitness in apparently
healthy adults: Guidance for prescribing exercise. Medicine and
Science in Sports and Exercise, 43, 1334–1359.
Gardner, B. (2012). Habit as automaticity, not frequency. European
Health Psychologist, 14, 32–36.
Gardner, B. (2014). A review and analysis of the use of ‘habit’ in
understanding, predicting and influencing health-related be-
haviour. Health Psychology Review. doi:10.1080/17437199.
2013.876238.
Gardner, B., Abraham, C., Lally, P., & de Bruijn, G. J. (2012).
Towards parsimony in habit measurement: Testing the conver-
gent and predictive validity of an automaticity subscale of the
Self-Report Habit Index. International Journal of Behavioral
Nutrition and Physical Activity, 9, 102.
Gardner, B., De Bruijn, G. J., & Lally, P. (2011). A systematic review
and meta-analysis of applications of the self-report habit index to
nutrition and physical activity behaviours. Annals of Behavioral
Medicine, 42, 174–187.
Gardner, B., & Tang, V. (2013). Reflecting on non-reflective action:
An exploratory think-aloud study of self-report habit measures.
British Journal of Health Psychology, 19(2), 258–273.
Godin, G., Jobin, J., & Bouillon, J. (1986a). Assessment of leisure
time exercise behavior by self-report: A concurrent validity
study. Evaluation De L’Exercise Physique Pendant Les Loisirs,
D’Apres Les Indications Fournies Par Les Interesses: une Eude
De Concordance, 77, 359–362.
Godin, G., Shephard, R. J., & Colantonio, A. (1986b). The cognitive
profile of those who intend to exercise but do not. Public Health
Reports, 101, 521–526.
Greenhouse, J. B., Stangl, D., & Bromberg, J. (1989). An introduction
to survival analysis: Statistical methods for analysis of clinical
trial data. Journal of Consulting and Clinical Psychology, 57,
536–544.
Greiner, M., Pfeiffer, D., & Smith, R. D. (2000). Principles and
practical application of the receiver-operating characteristic
analysis for diagnostic tests. Preventive Veterinary Medicine,
45, 23–41.
Grove, & Zillich. (2003). Conceptualisation and measurement of
habitual exercise. In Proceedings of the 38th Annual Conference
of the Australian Psychological Society (pp. 88–92). Melbourne:
Australian Psychological Society.
Hagger, M. S. (2010). Health psychology review: Advancing theory
and research in health psychology and behavioural medicine.
Health Psychology Review, 4, 1–5.
Hall, P. A., & Fong, G. T. (2007). Temporal self-regulation theory: A
model for individual health behavior. Health Psychology
Review, 1, 6–52.
123
662
Head, K. J., & Noar, S. M. (2014). Facilitating progress in health
behaviour theory development and modification: The reasoned
action approach as a case study. Health Psychology Review, 8,
34–52.
Hyde, A. L., Doerksen, S. E., Ribeiro, N. F., & Conroy, D. E. (2010).
The independence of implicit and explicit attitudes toward
physical activity: Introspective access and attitudinal concor-
dance. Psychology of Sport and Exercise, 11, 387–393.
IBM. (2011). IBM SPSS Statistics for Windows, Version 20.0.
Armonk, NY: IBM Corp.
Kasof, J. (2002). Indoor lighting preferences and bulimic behavior:
An individual differences approach. Personality and Individual
Differences, 32, 383–400.
Kaushal, N., & Rhodes, R. E. (2014). Exploring personality and
physical environment as predictors of exercise action control. In
Psychology of Extraversion Perspectives in Psychology Re-
search (pp. 91–105). New. York, NY: Nova Science Publishers.
Kraemer, H. C., Offord, D. R., Jensen, P. S., Kazdin, A. E., Kessler,
R. C., & Kupfer, D. J. (1999). Measuring the potency of risk
factors for clinical or policy significance. Psychological Meth-
ods, 4, 257–271.
Lally, P., & Gardner, B. (2013). Promoting habit formation. Health
Psychology Review, 7, S137–S158.
Lally, P., Van Jaarsveld, C. H. M., Potts, H. W. W., & Wardle, J. (2010).
How are habits formed: Modelling habit formation in the real
world. European Journal of Social Psychology, 40, 998–1009.
Linke, S. E., Robinson, C. J., & Pekmezi, D. (2014). Applying
psychological theories to promote healthy lifestyles. American
Journal of Lifestyle Medicine, 8, 4–14.
Lox, C. L., & Rudolph, D. L. (1994). The Subjective Exercise
Experiences Scale (SEES): Factorial validity and effects of acute
exercise. Journal of Social Behavior & Personality, 9, 837–844.
Luke, D. A., & Homan, S. M. (1998). Time and change: Using
survival analysis in clinical assessment and treatment evaluation.
Psychological Assessment, 10, 360–378.
Maddux, J. E. (1997). Habit, health, and happiness. Journal of Sport
and Exercise Psychology, 19, 331–346.
McAuley, E., & Courneya, K. S. (1994). The subjective exercise
experiences scale (SEES): Development and preliminary evalua-
tion. Journal of Sport & Exercise Psychology, 16, 163–177.
Moudon, A. V., Lee, C., Cheadle, A. D., Garvin, C., Johnson, D. B.,
Schmid, T. L., & Weathers, R. D. (2007). Attributes of
environments supporting walking. American Journal of Health
Promotion, 21, 448–459.
NIH. (2011). National Heart, Lung and Blood Institute from https://
www.nhlbi.nih.gov/guidelines/obesity/bmi_tbl.pdf. Retrieved
July 24, 2014.
Orbell, S., & Verplanken, B. (2010). The automatic component of
habit in health behavior: Habit as cue-contingent automaticity.
Health Psychology, 29, 374–383.
Ouellette, J. A., & Wood, W. (1998). Habit and intention in everyday
life: The multiple processes by which past behavior predicts
future behavior. Psychological Bulletin, 124, 54–74.
Rhodes, R. E. (2006). The built-in environment: The role of
personality and physical activity. Exercise and Sport Sciences
Reviews, 34, 83–88.
Rhodes, R. E. (2014a). Adding depth to the next generation of
physical activity models. Exercise and Sport Sciences Reviews,
42, 43–44.
Rhodes, R. E. (2014b). Will the new theories (and theoreticians!)
please stand up? A commentary on Sniehotta, Presseau and
Araújo-Soares. Health Psychology Review. doi:10.1080/
17437199.2014.882739.
Rhodes, R. E., Blanchard, C. M., & Matheson, D. H. (2006a). A
multicomponent model of the theory of planned behaviour.
British Journal of Health Psychology, 11, 119.
123
J Behav Med (2015) 38:652–663
Rhodes, R. E., Courneya, K. S., Blanchard, C. M., & Plotnikoff, R. C.
(2007). Prediction of leisure-time walking: An integration of
social cognitive, perceived environmental, and personality
factors. International Journal of Behavioral Nutrition and
Physical Activity, 4, 51. doi:10.1186/1479-5868-4-51.
Rhodes, R. E., & De Bruijn, G. J. (2010). Automatic and motivational
correlates of physical activity: Does intensity moderate the
relationship? Behavioral Medicine, 36, 44–52.
Rhodes, R. E., & De Bruijn, G. J. (2013). What predicts intention-
behavior discordance? A review of the action control framework.
Exercise and Sport Sciences Reviews, 41, 201–207.
Rhodes, R. E., Fiala, B., & Conner, M. (2009). A review and meta-
analysis of affective judgments and physical activity in adult
populations. Annals of Behavioral Medicine, 38, 180–204.
Rhodes, R. E., Matheson, D. H., & Blanchard, C. M. (2006b). Beyond
scale correspondence: A comparison of continuous open scaling
and fixed graded scaling when using social cognitive constructs
in the exercise domain. Measurement in Physical Education and
Exercise Science, 10, 13–39.
Rhodes, R. E., & Nasuti, G. (2011). Trends and changes in research
on the psychology of physical activity across 20 years: A
quantitative analysis of 10 journals. Preventive Medicine, 53,
17–23.
Rhodes, R. E., & Nigg, C. R. (2011). Advancing physical activity
theory: A review and future directions. Exercise and Sport
Sciences Reviews, 39, 113–119.
Rogers, W. (1974). Cognitive and physiological processes in fear
appeals and attitude change: A revised theory of protection
motivation. In J. T. Cacioppo & R. E. Petty (Eds.), Social
psychophysiology (pp. 153–176). New York: Guilford Press.
Rosenstock, I. (1974). The health belief model and preventive health
behavior. Health Education Monographs, 2, 354–386.
Rothman, A. J., Sheeran, P., & Wood, W. (2009). Reflective and
automatic processes in the initiation and maintenance of dietary
change. Annals of Behavioral Medicine, 38, S4–S17.
Salmon, J., Crawford, D., Owen, N., Bauman, A., & Sallis, J. F.
(2003). Physical activity and sedentary behavior: A population-
based study of barriers, enjoyment, and preference. Health
Psychology, 22, 178–188. doi:10.1037/0278-6133.22.2.178
Sentyrz, S. M., & Bushman, B. J. (1998). Mirror, mirror on the wall,
who’s the thinnest one of all? Effects of self-awareness on
consumption of full-fat, reduced-fat, and no-fat products.
Journal of Applied Psychology, 83, 944–949.
Sheeran, P., Gollwitzer, P. M., & Bargh, J. A. (2013). Nonconscious
processes and health. Health Psychology, 32, 460–473.
Shek, D. T. L., & Ma, C. M. S. (2011). Longitudinal data analyses
using linear mixed models in SPSS: Concepts, procedures and
illustrations. TheScientificWorldJournal, 11, 42–76.
Skinner, B. F. (1954). Science and human behavior. New York:
MacMillian.
Sniehotta, F. F., Presseau, J., & Araújo-Soares, V. (2014). Time to
retire the theory of planned behaviour. Health Psychology
Review, 8, 1–7.
Triandis, H. C. (1977). Interpersonal behavior. Monterey, CA:
Brooks/Cole.
Troiano, R. P., Berrigan, D., Dodd, K. W., Mâsse, L. C., Tilert, T., &
McDowell, M. (2008). Physical activity in the United States
measured by accelerometer. Medicine and Science in Sports and
Exercise, 40, 181–188.
Verplanken, B., & Aarts, H. (1999). Habit, attitude, and planned
behaviour: Is habit an empty construct or an interesting case of
goal-directed automaticity? European Review of Social Psy-
chology, 10, 101–134.
Verplanken, B., Aarts, H., & Van Knippenberg, A. (1997). Habit,
information acquisition, and the process of making travel mode
choices. European Journal of Social Psychology, 27, 539–560.
J Behav Med (2015) 38:652–663
Verplanken, B., & Melkevik, O. (2008). Predicting habit: The case of
physical exercise. Psychology of Sport and Exercise, 9, 15–26.
Verplanken, B., & Orbell, S. (2003). Reflections on past behavior: a
self-report index of habit strength. Journal of Applied Social
Psychology, 33, 1313–1330.
West, R. (2006). Theory of addiction. Oxford: Blackwell.
West, B. T. (2009). Analyzing longitudinal data with the linear mixed
models procedure in SPSS. Evaluation and the Health Profes-
sions, 32, 207–228.
WHO. (2015). Physical Activity from http://www.who.int/
dietphysicalactivity/pa/en/. Retrieved March 10, 2015.
Wiedemann, A. U., Gardner, B., Knoll, N., & Burkert, S. (2014).
Intrinsic rewards, fruit and vegetable consumption, and habit
strength: A three-wave study testing the associative-cybernetic
model. Applied Psychology: Health and Well-Being, 6, 119–134.
663
Williams, D. M., & Evans, D. R. (2014). Current emotion research in
health behavior science. Emotion Review, 6, 277–287.
Wood, W., & Neal, D. T. (2009). The habitual consumer. Journal of
Consumer Psychology, 19, 579–592.
Wood, W., Quinn, J. M., & Kashy, D. A. (2002). Habits in everyday
life: Thought, emotion, and action. Journal of Personality and
Social Psychology, 83, 1281–1297.
Warburton, D. E., Katzmarzyk, P. T., Rhodes, R. E., & Shephard, R.
J. (2007). Evidence-informed physical activity guidelines for
Canadian adults. Canadian journal of public health, 98, S16–
S68.
Zajonc, R. B. (1980). Feeling and thinking: Preferences need no
inferences. American Psychologist, 35, 151–175.
123